Power supply as a bottleneck for new AI data centers

29.04.2026

Source: E & M powernews

The demand for computing power for AI applications is growing faster than new grids and power plants can be built. This is attracting investors such as Blackrock.

The asset manager Blackrock wants to invest up to 100 billion euros in artificial intelligence. Around 75 to 80 percent is to flow into data centers, while 20 to 25 percent is earmarked for energy infrastructure, said Vibhu Kaushik from Global Infrastructure Partners. The company specializes in infrastructure financing and is part of Blackrock.

Kaushik was speaking at the Global Renewable Energy Summit, a corporate event organized by inverter and battery manufacturer Sungrow at its headquarters in Hefei, China. E&M was invited to this event as part of a press trip.

A key feature in the construction of future AI plant projects is the integration of energy generation and data centers. According to Kaushik, the two areas can no longer be considered separately. Energy is becoming a critical path for the implementation of new projects. Accordingly, the focus is shifting to solutions that enable the rapid provision of electricity.

According to Kaushik, "speed to power", i.e. the ability to provide a data center with sufficient power quickly, is becoming a decisive factor. In the coming years, the willingness to accept higher costs is therefore likely to increase if it means projects can be realized more quickly.

This is because there is an increasing discrepancy between the speed of the expansion of digital infrastructure and the traditional planning cycles of the energy sector. While energy suppliers develop and build their grids in five to seven years, the requirements of large technology companies for AI infrastructure are increasing much faster. "Companies such as Amazon, Google, Meta and Microsoft often plan their capacities in just twelve to 18 months."

"Access to energy is crucial"

According to Kaushik, this difference in speed is leading to a collision. The demand for power for AI applications is growing faster than new grids and power plants can be built. Location decisions for new data centers are usually based on access to space, (cooling) water or fiber optic connections. However, these factors are increasingly taking a back seat because: "Access to energy is becoming more and more important."

This is why his company has launched the "Artificial Intelligence Partners" (AIP) platform, which aims to mobilize capital specifically for the expansion of AI infrastructure. In addition to Blackrock and GIP, the initiative is backed by tech companies such as Microsoft and Nvidia as well as other partners with an investment volume of around 100 billion US dollars.

Decentralized concepts are playing a growing role in power supply. In addition to the traditional grid connection, data center operators are increasingly relying on behind-the-meter solutions. These include on-site generation plants such as gas turbines or engine power plants, often combined with battery storage and other components to stabilize the power supply.

This is intended to circumvent the often long connection times to the electricity grid. In a market as important as the USA, grid connection can take five to seven years. In order to implement projects more quickly, transitional solutions are therefore increasingly being developed in which data centers are initially operated independently and later connected to the grid.

High demands on power quality

According to Kaushik, there is no clear technological standard emerging. Rather, hybrid systems are emerging that combine different generation and storage technologies. In addition to traditional gas-fired power plants, experiments are also being conducted with fuel cells and other approaches. The decisive factor is less the individual technology and more the ability to "provide a quickly available and stable power supply".

With growth, the demands on power quality are also increasing. Short-term load fluctuations and high power densities in data centers require additional measures to stabilize the power supply. Battery storage and power electronics are therefore increasingly being seen as an integral part of the infrastructure and no longer just as backup systems.

Author: Stefan Sagmeister